0.10.0

Added

• Lesson 6 · Repo-aware AI assistant - a new working, polyglot lesson (Python, Node.js, and C#,
  byte-identical output). It indexes a small sample repo into line-numbered passages that each carry
  their own path:start-end citation, answers questions only from retrieved lines (always cited,
  or not found when nothing clears a score gate), and turns a change request into a
  plan-before-edit that changes no files. Ships with an interactive playground, an offline test,
  and a data/repo corpus + data/questions.json.
• Run the assistant on your own repository - every port honours a REPO_PATH environment variable
  and ask / plan subcommands (skipping .git, node_modules and build output). A ready
  extend/repo-ask wrapper and an EXTEND.md guide show three ways to take it further: a standalone
  CLI, a repo-search MCP tool you can register as a Claude Code skill (built on Lesson 2), and the
  port as a drop-in library.

Changed

• Lesson 4 shows the hardening more clearly - the prompt-injection demo and the interactive
  playground now label the two pipelines WITHOUT vs WITH hardening and tag each answer HIJACKED or
  SAFE, so the effect of the defences is obvious at a glance (Python, Node, and C#, byte-identical;
  toggle the defences in the playground and the WITH arm changes between SAFE, SAFE (leak blocked),
  and HIJACKED as the attack gets through or gets stopped).
• Dependencies - bump multer 2.1.1 → 2.2.0 in node/lesson-1 (Dependabot).